一种基于自适应子波神经网络的船舶噪声分类方法

An Efficient Engineering Classifier Capable of Automatically Extracting Features of Noise Signals Received

基于自适应子波神经网络提出了一种船舶噪声分类方法，构造了一种用于船舶噪声分类的自适应子波神经网络分类器，并应用该分类器对前置处理后的三类船舶噪声进行了自动地提取识别特征，并分类。该方法所获得的特征空间与以ＡＲ建模方法获得的特征空间相比，类别之间的可分性好，特征数目少，分类结果令人满意。

P.R. China′s fishery industry and offshore petroleum development have been in urgent need of a classifier of noise signals. After trying different lines of attack, we finally decided on the adaptive wavelet neural network shown in Fig.1. Then we derived the mathematical relationships, among which eqs.(4),(5) and (6) are the most important. Preprocessing is necessary before inputting signals into the adaptive wavelet neural network. Thus we succeeded in designing and implementing an efficint engineering classifier capable of automatically extracting features of noise signals received. The efficiency of our classifier can be clearly seen by comparing it with the neural network classifier whose feature vector is the AR parameter of signals. Fig.2 shows two-dimension (many dimensions reduced to two dimensions with well known method) features of AR classifier and Fig.3 shows those of our classifier. Figs.2 and 3 show clearly that our classifier is much better in discriminating between classes in feature space. Table 1 compares performances of our classifier with those of AR classifier. Among the five comparisons, the first and the last are most significant. The first comparison concerns network structure whose most important characteristics are number of input units and those of hidden and output units, which are all much less for our method. The last comparison concerns that the total average of our classifier is 98.9 and that the total average of AR classifier is only 88.7.